econstor Make Your Publications Visible.

Transcription

1 econstor Make Your Publications Visible. A Service of Wirtschaft Centre zbwleibniz-informationszentrum Economics Chiswick, Barry R.; Miller, Paul W. Working Paper International Migration and the Economics of Language IZA Discussion Paper, No Provided in Cooperation with: IZA Institute of Labor Economics Suggested Citation: Chiswick, Barry R.; Miller, Paul W. (2014) : International Migration and the Economics of Language, IZA Discussion Paper, No. 7880, Institute for the Study of Labor (IZA), Bonn This Version is available at: Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence.

2 DISCUSSION PAPER SERIES IZA DP No International Migration and the Economics of Language Barry R. Chiswick Paul W. Miller January 2014 Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor

3 International Migration and the Economics of Language Barry R. Chiswick George Washington University and IZA Paul W. Miller Curtin University and IZA Discussion Paper No January 2014 IZA P.O. Box Bonn Germany Phone: Fax: Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

4 IZA Discussion Paper No January 2014 ABSTRACT International Migration and the Economics of Language * This paper provides a review of the research on the economics of language as applied to international migration. Its primary focuses are on: (1) the effect of the language skills of an individual on the choice of destination among international (and internal) migrants, both in terms of the ease of obtaining proficiency in the destination language and access to linguistic enclaves, (2) the determinants of destination language proficiency among international migrants, based on a model (the three E s) of Exposure to the destination language in the origin and destination, Efficiency in the acquisition of destination language skills, and Economic incentives for acquiring this proficiency, (3) the consequences for immigrants of acquiring destination language proficiency, with an emphasis on labor market outcomes, and in particular earnings. Factors that are considered include age, education, gender, family structure, costs of migration, linguistic distance, duration in the destination, return migration, and ethnic enclaves, among others. Analyses are reported for the immigrant experiences in the US, Canada, Australia, the UK, Germany, Israel and Spain. JEL Classification: J15, J24, J31, J61 Keywords: immigrants, language, bilingualism, human capital, earnings Corresponding author: Barry R. Chiswick Department of Economics George Washington University 2115 G Street, NW Monroe Hall 340 Washington, DC USA brchis@uic.edu * Chapter prepared for Barry R. Chiswick and Paul W. Miller, eds. Handbook on the Economics of International Migration, Elsevier Inc (forthcoming). We appreciate the research assistance of Marina Gindelsky and Derby Voon.

5 International Migration and the Economics of Language 1. Introduction This chapter provides an overview of research on the economics of language, as applied to international migration. 1 The Economics of Language is the study of the determinants and consequences of language proficiency using the methodology and tools of economics. The beginning of interest by economists in language is usually attributed to Jacob Marschak (1965), who was concerned with the efficiency of communication. Using an evolutionary approach, those aspects of language that were beneficial in facilitating communication would survive, and those aspects (and languages) which were not efficient or effective would tend to disappear. In this framework languages tend to evolve over time. And just as mutations lead to the evolution of different species of plants and animals, isolation and language drift tend to promote the development of new dialects and languages. This approach did not generate much of a literature in economics on the evolution of languages. It did, however, stimulate interest in language usage in bilingual and multilingual countries (e.g., Canada, Switzerland). Which language or languages become dominant, especially for economic activity, and who learns which language? In some countries, regional or indigenous minority group languages have disappeared or appear to be on the verge of vanishing (e.g., consider Celtic in Ireland and Scotland, the Sami language in Sweden, and indigenous peoples languages in countries of overseas settlement such as the US, Canada, Australia, and Brazil). While in others, attempts have been made in recent decades with various degrees of 1 This chapter is a development of Barry R. Chiswick, The Economics of Language Learning for Immigrants: An Introduction and Overview in T.G. Wiley, J.S. Lee and R.W. Rumberger (eds.), The Education of Language Minority Immigrants in the United States, Bristol, UK: Multilingual Matters, 2009, pp

6 success to reinvigorate traditional languages (e.g., Catalonian in Spain and Welsh in Wales). Following several decades of low rates of international migration due to two world wars, the Great Depression, and severe immigration restrictions in many major destinations, international migration started increasing in the 1950s, and it has continued to increase in each decade since. This migration led to an influx of people in various destinations who were not familiar with the primary or dominant language of the destination. This led to interest by economists in the nexus between language and immigration. In addition to advances in economic theory and the development of testable models, economists now had better (not perfect, but better) micro data to estimate their models and test their hypotheses regarding language and international migration. Most of the research in the economics of language focuses on what can be described as microeconomics, that is, the behavior of individuals. The approach taken has been to view language skills as a form of human capital. The concept of human capital became important in the 1960s, with the emphasis on schooling, on-the-job training, health and information, all of which transform the person, and migration, which transforms the person s location (Schultz, 1962). It was only since the 1980s, however, that economists have viewed immigrant language skills as a form of human capital and analyzed it in this context (Carliner, 1981; McManus et al., 1983; Tainer, 1988). This interest arose as a result of the rapid growth of the non-english speaking portion of the increasing immigrant flows into the US and Canada, the emerging interest among economists in the determinants of the adjustment of immigrants to the host society, and the growing interest in the application of human capital theory (Becker, 1964). 4

7 Language skills satisfy the three requirements for human capital in that they are productive, costly to produce, and embodied in the person. First, a person s proficiency in the language of the area in which he or she lives is productive in the labor market. Those who speak/read the local language will find it easier to obtain a job and will generally be more productive on the job. In addition, language skills are productive in consumption activities. Those proficient in the local language will be more efficient in the search for higher quality goods and services and at lower prices. Any monolingual English speaker in the Chinese countryside quickly learns this proposition. Immigrants who do not speak the language of the broader society also find that their social and information networks are confined to their immigrant/linguistic enclave, rather than having a wider range. These benefits from proficiency provide economic and social incentives for immigrants to learn the host country s language. Second, acquiring language proficiency is not without costs. Immigrants spend a considerable amount of their own time and money (for language training schools, books, etc.) to become proficient in their new country s language. Acquiring language skills is not costless even for infants. Even if their own time has no economic value, the time of their parents or other caregivers in speaking and reading to the child is not costless. The costs involved in an immigrant learning a new language would be influenced by several factors, including the person s value of time (wage rate), the person s age, exposure to the destination language (as distinct from being able to avoid its use by living and working in a linguistic enclave) and the distance between the person s mother tongue and the language of the destination, among other factors. Finally, language skills are embodied in the person. Unlike owning physical capital (such 5

8 as a truck), but like learning to play a piano, language skills cannot be separated from the person. The idea that language skills are both productive and costly to acquire is not new, but rather at least thousands of years old. See Box 1, which relates the story of the Tower of Babel from the Biblical book of Genesis (Chapter 11, verses 1-9). The Tower of Babel provided a biblical explanation for the diversity of languages and the scattering of people: If, as one people with one language... then nothing that they may propose to do will be out of their reach. When their speech was confounded and they were scattered, they could no longer cooperate and they became less productive (Tanakh, 1985). 2 Box 1: Tower of Babel Everyone on earth had the same language and the same words. And as they migrated from the east, they came upon a valley in the land of Shinar and settled there. They said to one another, Come, let us make bricks and burn them hard. Brick served them as stone, and bitumen served them as mortar. And they said, Come let us build a city, and a tower with its top in the sky, to make a name for ourselves; else we shall be scattered all over the world. The LORD came down to look at the city and tower that man had built, and the LORD said, If, as one people with one language for all, this is how they have begun to act, then nothing that they may propose to do will be out of their reach. Let us, then, go down and confound their speech there, so that they shall not understand one another s speech. Thus the LORD scattered them from there over the face of the whole earth; and they stopped building the city. This is why it was called Babel, because there the LORD confounded the speech of the whole earth; and from there the LORD scattered them over the face of the whole earth. (Genesis, 11, 1-9) Source: Tanakh: The Holy Scriptures, pp (1986) Philadelphia, Jewish Publication Society. 2 An important issue currently facing the United States, and most of the highly developed economies, is the inverse of the Tower of Babel story. Immigration is resulting in the coming together of diverse peoples originally speaking a variety of languages who then merge over time into a common culture and a common language, even if they may also retain the languages of their origins. 6

9 The structure of this chapter is as follows. In the next section the three major aspects of the relation between language and international migration are outlined, and a broad overview of the methodology adopted in the empirical research is provided. This is followed by separate sections on each of the three major research themes, where the conceptual frameworks followed in the literature are presented, and empirical evidence discussed. The chapter ends with a summary of the findings and a discussion of gaps in the literature which warrant further research. 2. Research Issues and Methodology Much of the research on the economics of language as applied to international migration has focused on three main issues. The first matter addressed is the links between language background and the decision to migrate and the choice of destination by migrants. The language factors that are important include knowledge of a language that is used in the destination country, as well as knowledge of a language which, because it is linguistically close to a language used in the destination, makes learning the destination country language easier. The second issue concerns the determinants of proficiency in the primary or dominant language of the country of destination, including in the labor market; although the model and the methodology can be, and have been, applied to non-migrants who are linguistic minorities and native-born bilingual speakers. The third major concern covers the consequences for immigrants of obtaining proficiency in the dominant language. The consequences of language proficiency that has received the most attention have been in the labor market, particularly earnings. There are, however, other consequences which have not received much attention from economists or other social scientists. The primary focus on earnings has arisen in part because of interest in economic 7

10 well-being and in part because of the greater availability of data on earnings. Knowing the dominant language makes a person more efficient in the consumption of goods and services (higher quality and lower prices for goods and services). Investments in other forms of human capital, such as schooling and job training, are likely to be more productive if one can communicate in the dominant language in school and in the labor market. Knowing the dominant language of the destination can also increase the efficiency of parenting. Parents who are proficient in the dominant language can be more effective in teaching the language and culture of the destination to their children, which would be a benefit to them in school and later in the job market. Language skills also have social benefits as they can expand the range of friendship networks beyond one s ethnic/linguistic enclave. Finally, civic involvement is enhanced with knowledge of the host country s language. Indeed, for the United States and many other countries, at least a basic knowledge of the destination language is required for immigrants to become citizens and acquire full political and economic rights. This brings about increased political empowerment. There is no doubt that these non-labor market consequences of dominant language proficiency are important. However, to have a manageable review we focus on the labor market consequences. The particular models advanced by economists in relation to each of these research questions are discussed in Sections 3, 4 and 5. The testing of the models, or the estimation of the equations, relies on multivariate statistical (econometric) techniques. In general, many economists believe in the importance of testing for the robustness of findings. One set of estimates from one data set may be insightful, but cannot determine whether the results are unique to that data, group, country or time period, or whether they are 8

11 generalizable across these dimensions. A hypothesis or model that is not robust, but is valid for only a unique group, time and place, is clearly of very limited value. On the other hand, one has much greater confidence in a hypothesis or model that is robust, that is, supported by analyses of diverse data sets. The analyses reported below represent a synthesis of the findings on immigrants for different types of data, censuses and surveys, both cross sectional and longitudinal. They are for immigrants who have legal status, as well as those with an illegal or unauthorized status. Although the data analyses reported below are for the late 20th and early 21st centuries, they are for different data sets across several countries, primarily the United States, Australia, Canada, Germany and Israel, where the destination language is English in the first two, English and French in Canada, German in Germany, and Hebrew in Israel. The particular value of research on Israel and Germany in this context is that whereas English, and to a lesser extent French, is an international language of culture, business and science, which is often learned in school in the country of origin as a second language, this is less so for Hebrew for immigrants to Israel, or for German among the groups of immigrants covered in the empirical research for Germany. That the findings for Israel and Germany parallel those of the other countries is a test of the robustness of the model across destination languages (Chiswick, 1998; Chiswick and Repetto, 2001; Dustmann, 1994). In the case of the links between language background and destination language choice, many destination countries, most countries of origin in the world, and a wide range of languages, are covered. There are several dimensions of language skills oral (speaking and hearing) and literacy (reading and writing). Survey and census data on the language skills of immigrants almost 9

12 always rely on self-reported responses or responses provided by an adult household member. Although some data sets report responses to questions for immigrants on reading and writing, most of the data are regarding speaking skills, focusing on either the self-reported level of competency or identifying the languages spoken on a regular basis. Analyses using literacy skills show the same patterns as those using speaking skills, in part because the two are so highly correlated (Chiswick, 1991; Chiswick and Repetto 2001; Dustmann, 1994). The discussion here will be expressed in terms of speaking proficiency, unless noted otherwise. 3. Choice of Destination Migration, whether internal to a country or international, is an investment in human capital. Unlike other forms of human capital, such as schooling, health and information which transform the person, migration transforms where the person lives or works. It is a form of human capital because migration is costly, is beneficial in either consumption or production (e.g., labor market work), and the migration per se cannot be separated from the person. The costs of migration include out-of-pocket costs (sometimes referred to as direct costs) and foregone earnings (sometimes referred to as opportunity costs). The costs are far greater than merely the costs of moving oneself, family members, and household goods from one point (the origin) to another (the destination). The costs incurred in the origin include the separation from family, friends and a familiar environment. The set of skills acquired in the origin, including language skills, and which are useful in consumption and production activities, may not be equally useful in the destination. The transferability from the origin to the destination may be limited by geographic differences in technology, by custom, by occupational licensing, etc. The 10

13 transferability may also be limited by language differences, where language can be considered a technology for communication. A Chinese speaker in a monolingual English country would find that being a consumer who looks for higher quality goods and services at lower prices is more difficult than it was in China, or that the job search process is more difficult. Wages would be lower and employment conditions less desirable if linguistic disadvantages lower the workers productivity on the job or increase workplace costs (e.g., accidents). Relevant language skills may be important for communicating with supervisors, peers, and subordinates, as well as suppliers and customers or clients. As a result, part of the cost of migration is the lower earnings during the period of adjusting one s language skills, as well as the cost of direct investment in improving destination language skills (e.g., the opportunity cost and out-of-pocket cost of a language training program). The latter cost may be incurred prior to the migration, in anticipation of the move, or after migration. The costs of language adjustment depend on many factors, which will be discussed in greater detail below. These costs will include the importance and the ease or difficulty of learning the language of the destination. Potential migrants need not consider just one destination, but may consider the range of potential or available destinations. Therefore, part of the calculation as to which destination is most preferred is the cost of destination language acquisition. The language acquisition costs are lower the closer is the language of the origin to that of the destination. Thus, for an Australian, the linguistic cost of adjusting to a destination is cheaper if it is the UK rather than France, or for a Spaniard if the destination is Argentina, rather than Canada. The implication is that language differences are among the factors that influence 11

14 the choice of destination. Much of the research on the links between language background and the choice of destination has used aggregate-level data. Two decisions have been studied: the choice of initial destination country; and internal migration in the years following arrival in the host country. Evidence from both streams of literature is reviewed. 3.1 Choice of initial destination country The research on the links between language background and choice of the initial destination can be illustrated through detailed coverage of the study by Clark, Hatton and Williamson (2007), and then covering other studies more briefly. Clark et al. (2007) try to account for the determinants of migration rates to the US by place of birth for 81 source countries from 1971 to They based their analysis around the following model: 3 mig pop = f ( y y, syr syr, inq inq, age, pov, dist, land, eng, j j j US j US j US j j j j j stock pop, policy variables) j j where mig j pop j is the flow of migrants from source country j to the US in a particular year, normalized by the population of the source country, yj y US is the average (purchasing power parity adjusted) income in the source country j relative to that in the US, the terms in syr and inq capture, respectively, differences in average years of schooling and inequality in source country j and the US, age is the share of the population in the sending country aged 15-29, pov is 3 Time subscripts are suppressed to simplify the notation. 12

15 the poverty rate in the origin country, dist is the geographic distance of the source country from the US, land denotes cases where the origin country is land-locked, eng is for where the source country is predominately English speaking, and the stock variable captures the number of previous immigrants from the source country. The Clark et al. (2007) model also contained policy variables, for the number of visas available in the different visa classes, and institutional factors, such as the US 1986 Immigration Reform and Control Act (IRCA) legalization program. Hence, it is seen that the model of equation (1) captures economic and demographic drivers of migration in the first five terms, with the remaining terms representing costs and policy parameters. In particular, it was argued that the cost of migration to the US would be higher if the sending country was not predominantly English speaking. Clark et al. s (2007) results show that the English-speaking sending countries were associated with a statistically significant higher migration rate to the US in the models that did not include the immigrant stock variables. In other words, having an English language background matters when it comes to understanding rates of migration to an English-speaking country. This effect, however, was not statistically significant where the stock of previous migrants from the sending country was included in the estimating equation. As noted by Clark et al. (2007. p.267) Since the immigrant stock reflects past immigration, it captures much of the effect of slow moving fundamentals over the longer term. Included in these fundamentals would be the country-specific cost factors that the English-speaking background variable is used to represent. Approaches similar to the model that Clark et al. (2007) applied to English-speaking countries have been adopted by other researchers when considering migration flows into 13

16 countries characterized by a number of official languages. For example, Karemera, Oguledo and Davis (2000) examine migration flows to the US, using a common language (English) variable along the lines of Clark et al. (2007), and to Canada, where they use a variable for the language of the origin country being either English or French. Karemera et al. (2000) report, however, that language commonality was not a significant determinant of migration rates to either the US or Canada over the decade One of the main differences between the models of Karemera et al. (2000) and Clark et al. (2007) is that the former includes a set of dichotomous variables for region of origin. 4 The coverage of common languages has been extended further by Pedersen, Pytlikova and Smith (2008). They examined gross migration flows from 129 countries into 22 OECD destination countries over The 22 OECD countries have a number of official languages (English, German, Spanish, Italian, etc.), and so the English language variable in the model of Clark et al. (2007) was replaced by a common language variable in the Pedersen et al. (2008) study. This variable was set equal to one where there is a common language between the origin and destination, and it is set equal to zero where there is no common language, with the Ethnologue: Languages of the World (2009) being used to classify pairs of countries (see Box 2). An additional feature of this study is that it also includes a dummy variable for countries that were ever in a colonial relationship. Presumably this variable is correlated with the common language variable, though this was not examined in the study. Nevertheless, Pedersen et al. (2008) report that the common language variable was an important influence on migration flows, 4 The studies in this field are characterized by differences in the choice of dependent variable (whether the immigrant flow is normalized by the source country population, and whether flows or stocks are used), the selection of independent variables, as well as the lag structure of variables. We do not discuss these specification issues in detail here. 14

17 being significant in five of the eight models presented in their main set of results. Box 2: Ethnologue: Languages of the World (16 th edition of 2009) This is a comprehensive reference volume that catalogues 7,413 languages, including details on the 6,909 known living languages in the world at the time of writing. Each language is part of a language family (that is, its linguistic lineage is provided). For example, the linguistic lineage for English is, from largest grouping to smallest, Indo-European Germanic Germanic West English. This type of connectivity between languages has been used in various ways by researchers to construct a measure of linguistic distance or proximity. It can be illustrated using the algorithm proposed by Adsera and Pytlikova (2012). Thus, they construct their variable as follows: First we defined weights: the first equal to 0.1 if two languages are related at the most aggregated linguistic tree level, e.g. Indo-European versus Uralic (Finnish, Estonian, Hungarian); the second equal to 0.15 if two languages belong to the same second-linguistic tree level, e.g. Germanic versus Slavic languages; the third equal to 0.20 if two languages belong to the same thirdlinguistic tree level, e.g. Germanic West vs. Germanic North languages; and the fourth equal to 0.25 if both languages belong to the same fourth level of a linguistic tree family, e.g. Scandinavian West (Icelandic) vs. Scandinavian East (Danish, Norwegian and Swedish), German vs. English, or Italo-Western (Italian, French, Spanish, Catalan and Portuguese) vs. Romance Eastern (Romanian). Then, we constructed the linguistic proximity index as a sum of those four weights, and we set the index equal to 0 if two languages did not belong to any common language family, and equal to 1 if the two countries had a common language. Thus the linguistic proximity index equals 0.1 if two languages are only related at the most aggregated linguistic tree level, e.g. Indo-European languages; 0.25 if two languages belong to the same first- and secondlinguistic tree level, e.g. Germanic languages; 0.45 if two languages share the same first- up to third-linguistic tree level, e.g. Germanic North languages; and 0.7 if both languages share all four levels of a linguistic tree family, e.g. Scandinavian East (Danish, Norwegian and Swedish). (Adsera and Pytlikova, 2012, p.12). Obviously, other scales can be derived using this information, see for example, Belot and Hatton (2012), and this has been argued by Isphording and Otten (2012) to be a weakness of the approach. They argue, for example This linguistic-tree approach has to deal with strong cardinality assumptions, and arbitrarily chosen parameters. Additionally, the approach offers only low variability between different language pairs and is difficult to implement for isolated languages such as Korean (Isphording and Otten, 2012, p.5). Both Adsera and Pytlikova, (2012) and Belot and Hatton (2012) have used these data for measures of linguistic distance in immigration research. Other researchers, for example Pedersen et al. (2008), have used this source to compile a common language dummy variable. Source: Lewis, M. P. (ed.), (2009). Ethnologue: Languages of the World, Sixteenth edition. Dallas, Tex.: SIL International. Online version: 15

18 Mayda (2010) and Ortega and Giovanni (2009) also examine the determinants of immigration flows to OECD countries. Mayda (2010) covered 14 OECD countries, 79 sending countries, and the time period , whereas Ortega and Giovanni (2009) cover 14 OECD countries, 73 countries of origin, and the longer time period of 1980 to Neither study found language background to be of importance to the explanation of migration flows. Ortega and Giovanni (2009, p.14) argued that This is hardly surprising as most of the large migratory flows to the OECD (except for Mexico-US) take place between countries that do not share a land border or a common language. Moreover, Mayda (2010, p.1263) notes that The impact of a common language, though of the right sign, is not statistically significant and, surprisingly, past colonial relationships do not appear to affect migration rates (this is true whether common language and colony are entered in the regression together or one at a time). A more recent study that reports that a common language is important to understanding international migration is Grogger and Hanson (2011). They study the stock of immigrants in 15 high-income OECD countries in 2000, and employ both common language and an Englishspeaking destination country variables. Their model has separate equations for the scale of international migration flows of low-skilled (primary educated) and high-skilled (tertiary educated) workers, for the selection on the basis of the skills of immigrants (primary or tertiary educated from a particular origin country in the destination, compared to those who remained in the origin) and for sorting across destinations (the mix of immigrants across the destination countries). The preferred set of results for the scale equation shows that immigrants are more likely to be from a country that has a language in common with the destination country. The results for the selection equation indicate that immigrants that move to a country with a language 16

19 in common with the origin country are positively selected in terms of education levels, while the findings for the sorting equation suggest that destinations that have a language in common with the origin country attract more highly skilled immigrants. Similar findings are reported for the English-speaking destination country variable. Beine et al. (2011) examined matters similar to Grogger and Hanson (2011), but from the perspective of changes in immigrant stocks between 1990 and 2000, using data on 195 source countries and 30 OECD countries. Their common language variable was a highly significant determinant of both low-skill and high-skill migration flows. A common language was also associated with a statistically significant positive effect on the skill ratio of these migration flows. Hence, while the results from the research that has been based on the importance of a common language to understanding international immigration are somewhat mixed, the findings predominately suggest that a common language is an important determinant of the scale and mix of the migration flows. There have been a number of developments in the analyses of the effect of language on the choice of destination. Of greatest relevance to this review are the studies that use a measure of linguistic distance in place of the common language variable. This development reflects the fact that migration costs are lower where the migrant can easily learn the language of the destination country, for example, for migrants with a mother tongue that is linguistically close to the dominant language of the destination (see the next section). Belot and Hatton (2012) examine the characteristics of migration for 70 source countries and 21 OECD destination countries for 2000/2001. While this study continues the theme of the 17

20 research by Grogger and Hanson (2011) and Beine et al. (2011) by focusing on educational selectivity, the key feature from our perspective is their linguistic proximity variable. They derive this measure from the language family information presented in the Ethnologue: Languages of the World (See Box 2). As constructed by these authors, the linguistic proximity variable has values from 1 to 5, according to the number of common nodes in the linguistic tree between the closest official languages of pairs of countries. The Ethnologue common node measure of Belot and Hatton (2012) was statistically significant and positive in their skill selection equation, a finding that the authors argue shows that the transferability of human capital might be easier when the linguistic gap is less, and hence immigrants can readily learn the dominant language of the destination. 5 There might be a suspicion that the stronger results obtained with the seemingly superior measure of linguistic distance in Belot and Hatton (2012) is linked to their focus on stocks of immigrants. After all, Grogger and Hanson (2011) found that the conventional common language variable was highly significant in their model that has a focus on stocks, rather than annual flows. This matter can be addressed through reviewing the research of Belot and Ederveen (2012). They examine a panel of 22 OECD countries for the period The measure of linguistic distance used in their study was based on the work of Dyen et al. (1992): Dyen et al. (1992) constructed a measure of the distance between Indo-European languages based on the proximity of 200 words from each language (See Box 3). This variable was included in the estimating equations along with a common language variable. Unlike the research of Belot and Hatton (2012), both the common language and linguistic distance variables were statistically 5 Belot and Hatton (2012) noted that they also estimated equations that included a dichotomous common language variable, and that this common language variable was insignificant in the presence of the linguistic proximity variable. 18

21 significant, although the impact of sharing a common language dropped by over one-quarter when the measure of linguistic distance was included in the estimating equation. Box 3: The Dyen Lexicostatistical Percentage Approach Comparative lexicostatistics is the study of historical relations among speech varieties belonging to the same language family through a quantitative study of cognation among their vocabularies. The lexicostatistical percentage approach is the oldest and most widely-used lexicostatistical approach. As explained by Dyen et al. (1992), their application of this approach has four phases. First, they worked with the 200 meanings that had been proposed by Swadesh, and developed phonetic representations (forms) of the words with these particular meanings for the chosen languages (see Swadesh (1952) for an earlier compilation). Then the cognation among the forms in two languages was established through expert judgment. Cognation requires that the forms have descended in unbroken lines from a common ancestor in the same language family. Consideration of the number of cognate forms from the list of 200 meanings gives rise to the so-called lexicostatistical percentage. For example, the value when German and English are compared is 57.8 percent. The value for the French-English comparison is 23.6 percent. In other words, German and English are more similar than French and English. The final phase of work by Dyen et al. (1992) involved the categorization of the languages into various groups. It is to be noted that the Dyen et al. work covers only Indo-European languages. An example of Dyen et al. s (1992) numbers is given below. Dyen Matrix of Linguistic Distances (higher values mean smaller distance) Languages Languages Italian French Spanish German Dutch Danish English Greek Italian French Spanish German Dutch Danish English Greek Examples of economics studies using these data are Ginsburgh, Ortuño-Ortín and Weber (2005), Belot and Ederveen (2012) and Adresa and Pytlikova (2012). Source: Dyen, I., Kruskal J.B. and Black, P., (1992). An IndoEuropean Classification: A Lexicostatistical Experiment, Transactions of the American Philosophical Society, New Series 82(5). 19

22 An idea of the relative importance of the Belot and Ederveen (2012) measure of linguistic distance for migrant flows can be found from the effects that these authors computed for one standard deviation increases in the various explanatory variables. They report that Our regression results imply that an increase in linguistic distance with one standard deviation lowers the migration flow with 56%...This effect is about 50% higher than the effect of raising GDP per capita in the destination country by one standard deviation and much more than a change of one standard deviation in unemployment rates (Belot and Ederveen, 2012, p.1096). The importance of the linguistic distance measure in the analysis of migration flows was found in the many tests of robustness these authors conducted. The final study in our review is by Adsera and Pytlikova (2012). They cover immigration flows in 30 OECD countries from 233 source countries, for the years Their research is important in the study of the links between linguistic distance and the destination choice of immigrants because they used many measures of linguistic distance or linguistic proximity. Their preferred measure was based on the Ethnologue: Language of the World, and ranges from zero to one according to the number of levels of the language family tree shared by the destination and source country languages (See Box 2). Belot and Hatton (2012) also use this type of measure. In addition, in tests of robustness, Adsera and Pytlikova (2012) use both a measure based on Dyen et al. (1992) (see also Belot and Ederveen (2012)) and a measure based on the Levenshtein linguistic distance approach produced by the Max Planck Institute for Evolutionary Anthropology. See Box 4. 20

23 Adsera and Pytlikova (2012) report that their preferred measure of linguistic proximity is a statistically significant determinant of migration flows. 6 This holds in both bivariate and various multivariate models, including those that take account of both the stock of migrants in the destination and the flow of migrants between countries in the previous period, as well as in models that include destination and origin country fixed effects. Moreover, the finding that linguistic proximity is an important determinant of migration flows was robust with respect to the use of the two alternative measures of linguistic proximity (based on Dyen et al. (1992) see Box 3; and the Max Planck Institute for Evolutionary Anthropology see Box 4). It was also robust with respect to the choice of language to use in the construction of the proximity measure (main official language, any official language, and the major language, where major was defined as that which was used most extensively). It was also robust with respect to when the effects are estimated separately for English-speaking countries and for non-english-speaking destination countries. The estimated impacts were, however, stronger for non-english-speaking destination countries. Adsera and Pytlikova (2012, p.25) argue that The likely higher proficiency of the average migrant in English rather than in other languages may diminish the relevance of the linguistic proximity indicators to English speaking destinations. This greater proficiency in English is likely due to English having become the international language of science, technology, and business. 6 Adsera and Pytlikova (2012) also used measures of the diversity of languages in both the country of origin and the country of destination. 21

24 Box 4: The Max Planck Institute for Evolutionary Anthropology measure of Levenshtein linguistic distance The Max Planck Institute for Evolutionary Anthropology in Germany has used a lexicostatistical approach to develop a measure of linguistic distance, or more precisely the Levenshtein distance, using an algorithm that compares pronunciation and vocabulary of language pairs. This procedure is based on the Automatic Similarity Judgment Program (ASJP). The starting point for this approach is a small Swadesh list (see Box 3) of 40 words that describe common things and environments. These words are then expressed in a special phonetic transcription known as the ASJP code. This code uses 41 characters on a standard QWERTY keyboard to represent the common sounds in human communication. Then the number of additions or subtractions of characters (or sounds) required to transform a word in one language into the same word in another language is computed, using the ASJP. Isphording and Otten (2012, p.7) offer the following illustration: to transfer the phonetic transcription of the English word you, transcribed as yu, into the transcription of the respective German word du, one simply has to substitute the first consonant. But to transfer manunt3n, which is the transcription of mountain, into berk, which is the transcription of the German Berg, one has to remove or substitute each 8 consonants and vowels, respectively. This evaluation is then adjusted to account for differences in world length and the potential similarities in phonetic inventories that might lead to similarity by chance to give the Levenshtein distance measure. Larger values thus indicate languages that are further apart. Some examples of this measure of linguistic distance are provided below. Examples of Closest and Furthest Language Pairs with Respect of the Levenshtein Distance Measure Closest Furthest Language Distance Language Distance Distance to English Afrikaans Vietnamese Dutch Turkmen Norwegian Hakka (China) Distance to Spanish Galician Wolof (Senegal) Italian Igbo Onitsha (Nigeria) Portuguese Ewondo (Cameroon) Source: Extracted from Table 1 in Isphording and Otten (2012). This data source has been used in economic research by Isphording and Otten (2011)(2012) and Adsera and Pytlikova (2012). 22

25 Table 1 summarizes the evidence on the links between language background and the choice of destination among immigrants. It shows that language background matters to destination choice, and that stronger empirical results emerge in studies that use measures of linguistic distance than in studies that employ simple dichotomous variables to reflect a common language between pairs of countries. In other words, it is not just the knowledge of a destination dominant language that matters, but the ease with which an immigrant can learn the destination dominant language is also very important. Table 1: Overview of Studies into the Links Between Language Background and Destination Choice of Immigrants Does language matter to destination Authors Countries studied and time period Language variable choice? Clark, Hatton and Williamson (2007) Flows to the US from 81 source countries, English dummy Yes Karemera, Oguledo and Davis (2000) Pedersen, Pytlikova and Smith (2008) Mayda (2010) Ortega and Giovanni (2009) Grogger and Hanson (2011) Beine, Docquier and Özden (2011) Flows to US and Canada, Flows to 22 OECD countries from 129 source countries, Flows to 14 OECD countries from 79 sending countries, Flows to 14 OECD countries from 73 source countries, Stock of immigrants in 15 highincome OECD countries in 2000 Change in stocks between 1990 and 2000, for 30 OECD countries, with 195 source countries Belot and Hatton (2012) Stock of immigrants from 70 source countries in 21 OECD countries, Belot and Ederveen (2012) Flow of immigrants to 22 OECD countries, English/English or French dummy Common language dummy Common language dummy Common language dummy Common language dummy Common language dummy Based on the Ethnologue (see Box 2) Based on Dyen et al. (1992) (see Box 3) No Yes No No Yes Yes Yes Yes

26 Adsera and Pytlikova (2012) Flows of immigrants to 30 OECD countries from 233 source countries, Based on the Ethnologue (see Text Box 2), Dyen et al. (1992) (see Box 3) and the Levenshtein distance approach (see Box 4) Yes 3.2 Location choice within a country The seminal study on the location choice of immigrants within a country is Bartel (1989) which used 1980 Census data. 7 This covered the US. The main set of results in this paper was from a conditional logit model, estimated separately for Asians, Hispanics and Europeans. The explanatory variables used in Bartel s (1989) model included characteristics of the areas within which the immigrants lived, such as the unemployment rate, average wage, and the distance of the location from the immigrant s country of origin. The main such variable, however, was the ethnic concentration measure, defined as the percentage of a specific ethnic group that resided in the particular location. The ethnic concentration variable was a highly significant determinant of location choice for each ethnic group, and within each ethnic group, for the three arrival groups considered. This effect tended to be weaker among the more educated immigrants. Using the Census information on place of residence five years ago, Bartel (1989) examined internal migration patterns. Immigrants were reported to be more likely to change locations than the native born, and much of this movement was associated with an increase in the geographical concentration of the ethnic group. While this pioneering study did not examine the role of 7 There is no difference in the theory of migration between internal and international migration. Institutional factors, such as regulations, may differ. 24

27 language background per se, the apparent links between ethnic concentrations and language backgrounds were developed in subsequent research. 8 The approach of Bartel (1989) has been developed by Jaeger (2000), among others. Jaeger (2000) used data from the Immigration and Naturalization Service on immigrants admitted to the US during the fiscal year, combined with data from the 1980 and 1990 censuses. Bartel s (1989) ethnic concentration variable was expanded to consider the immigrant s region of birth, and the share of immigrants who speak a language other than English that is spoken in the immigrant s country of birth. These location characteristics were considered in conjunction with the share of the population in the location that was born abroad. Jaeger (2000, p.15) reported that Region-of-birth concentrations are about 3 times as important in determining location as language and foreign-born shares. This relative importance held for all visa types other than for the small group of diversity visa immigrants. It also carried over to the analysis of the location choice of the foreign born in the US who received an adjustment of state to become permanent resident aliens. Nevertheless, this research showed that language background matters for the location choice of immigrants within a country. The research by Bauer, Epstein and Gang (2005) is focussed on the links between the location choice of immigrants from Mexico in the US and their English language proficiency. Specifically, they ask whether the choice of the size of enclave community in which to settle is affected by English language proficiency. The analyses were based on data from the Mexican 8 Zavodny (1999) reported that the fraction of the state population that is foreign born was a highly significant determinant of recent immigrants location choice for all admission categories considered in her study (family, employment, Immigration Reform and Control Act of 1986 (IRCA) conversions to legal status, and refugee/asylee conversions to legal status), and for all country groups examined (Chinese, Dominican Republic, Mexico, Philippines, Vietnam). 25